[CMake] [PATCH] slightly modify cache variable behaviour, Was: Re: option() behavior

[Bill Hoffman]

Alexander Neundorf wrote:
> On Monday 11 May 2009, Bill Hoffman wrote:
>> So, CMake has done what it does now from the start.   There was a short
>> period of time when it did not, and that was when a re-write was done,
>> and it quickly broke some existing projects.  Here is the thread:
> 
> Yes, I can imagine that.
> Still the current behaviour is not very intuitive, especially that set(CACHE) 
> in the first cmake run overrides the "visible" variable, while the set(CACHE) 
> in the second run basically does nothing, and so the value of the "visible" 
> variable is different.
> I think for CMake 2.8 or 3.0 it would be a good idea to make this more 
> consistent and easier to understand (especially since now more and more 
> people are using cmake).
> 
>> http://www.cmake.org/pipermail/cmake/2007-March/013204.html
>>
>> I think the rule should be if you want to change a variable you have to
>> have the set AFTER the variable, or use CACHE INTERNAL.
> 
> What do you mean exactly with "have the set after the variable" ?
> Have the set(FOO <value>) after the set(FOO <value> CACHE), or something 
> else ?
> 

Yes,

This always has worked as expected:

set(FOO 2 CACHE)
set(FOO 1)  # this is always the value

However, I think your change is OK.  Basically if the set CACHE comes 
after the set, it will always set the value.   We do need a policy for 
this, but it should be easy to detected.  You will have to see if the 
REMOVE is going to do anything.    I wonder how many warnings that 
policy will cause... I  guess we could create the policy code, and try 
it on VTK, Boost, SecondLife, and some other big projects to see what 
happens...

-Bill